Anthropic wants Australia to let it train AI on copyrighted work, and the ABC says that will cannibalise the news

Anthropic has asked the Australian government to let AI companies train their models on copyrighted Australian work, as long as they meet a set of conditions, in a submission to parliament’s Joint Select Committee on Artificial Intelligence. Australia’s two public broadcasters, the ABC and SBS, have told the same inquiry the opposite: AI companies should pay for the news they use.

The submissions, among more than 500 the committee has published, land days before Anthropic and OpenAI executives appear before the committee in Sydney on Tuesday, Oct 6, where OpenAI will also face questions over its agent breaking into the Medicare portal.

“A narrow form of conditional approval”

Australia’s government has already ruled out a text and data mining exception, the kind of broad carve-out that would let AI firms train on copyrighted material freely. Anthropic says it accepts that, and proposes a narrower route instead:

As an alternative, we believe that the Government could permit AI model training through a narrow form of conditional approval.

Anthropic, submission to the Joint Select Committee on Artificial Intelligence

Under its plan, training on material that could be under copyright would be approved for companies that meet the conditions. Those could include respecting “technical protection measures or ‘do not crawl’ signals through robots.txt”, the file websites use to tell bots what they can and can’t read, and meeting a proposed set of AI standards. Anthropic says the government “may choose to require investment or other conditions designed to support the future of Australian creators”, and that it is “open to considering those conditions”.

The submission pitches this as Australia forging “its own system to provide legal certainty under copyright that brings AI developers to the table”. It ties it to Anthropic’s plans in the country, including its investment in the Western Downs Digital Park data centre site in Queensland, and calls for Australia to support “frontier model training” there.

“Cannibalisation” of the news

The broadcasters see it very differently. In its submission, the ABC said AI services were treated more favourably than Australian media, the Guardian reports:

This unequal treatment will accelerate the cannibalisation of the Australian news industry.

ABC, submission to the Joint Select Committee on Artificial Intelligence

SBS rejected the idea that robots.txt protects anyone, saying such mechanisms “are routinely ignored or bypassed” and that “the onus should be on AI companies to avoid wrongdoing”. Both broadcasters want AI companies brought under the news bargaining incentive, which pushes big platforms such as Google and Meta into paying deals with news outlets, and the ABC wants AI services held to the same copyright, defamation and privacy rules as publishers.

Safety pitch alongside

The rest of Anthropic’s submission mixes the economic case with warnings. It says AI could add up to $116 billion (about US$77 billion) to Australia’s economy over a decade, citing the Productivity Commission, but has “serious implications for Australia’s national security”. It repeats Dario Amodei’s call to “pace the frontier” and says Anthropic will embed a permanent team of independent outside evaluators, free to publish their findings.

Why it matters

Copyright fights over AI training are being settled country by country, and courts have started to go against AI firms, as in the Thomson Reuters v Ross appeal. An opt-out system built on robots.txt would put the burden on publishers to block AI, which is exactly what the ABC and SBS are fighting, and Australia’s choice will be watched by other governments facing the same pressure.

Sources: Anthropic, submission 305, Joint Select Committee on Artificial Intelligence submissions; The Guardian (ABC and SBS submissions).